As this study relied exclusively on publicly available data without human participants, it was reviewed by the Oregon Health & Science University Institutional Review Board and deemed not to be human research (00027957).

This cross-sectional content analysis aimed to evaluate the quality, accuracy, and attitudes of popular sunscreen content on TikTok.

To establish a search strategy, 2 reviewers (KD and JR-M) first conducted exploratory searches using targeted terms such as “toxic sunscreen,” “sunscreen risks,” and “sunscreen myths.” On the basis of these findings, we selected the broad search term “sunscreen,” limiting our final dataset to a single, neutral term to minimize search bias and ensure broad inclusion of relevant content. This approach enabled capturing a wide range of perspectives within a single organic search.

The final search was conducted on August 14, 2024, using newly created TikTok accounts to reduce algorithmic personalization. We applied TikTok’s “most-liked” filter and identified the top 106 most-liked videos returned by the search term “sunscreen.” For each video, we recorded metadata (URL, creator username, date posted, number of likes, views, and creator follower count) in Microsoft Excel (Microsoft Corporation) at the time of search.

The top 100 most-liked videos were screened for inclusion. Eligible videos were required to (1) be in English or contain English subtitles and (2) be relevant to sunscreen use, safety, or recommendations. Duplicates and irrelevant videos were excluded. Videos ranked 101 to 106 were used exclusively for coder calibration and interrater reliability (IRR) testing. A sample size of 100 was selected to balance feasibility of detailed manual coding with representativeness. Although larger-scale TikTok analyses exist, such as the study by Steinke et al [9], our aim was to balance depth with representativeness.

Videos were coded for the following variables:

Two reviewers (KD and JR-M) independently coded 6 calibration videos (videos 101-106) to refine the structured codebook (Multimedia Appendix 1) and assess IRR. IRR for categorical variables was evaluated using percent agreement, Scott pi, Cohen kappa, and Krippendorff alpha. Using the original 1‐5 GQS wording from prior literature [10,12-14], initial GQS agreement was 60% despite more than 85% agreement on other variables.

To improve consistency, coders received additional training and in-person calibration meetings, clarified the construct to reflect how lay viewers interpret scientific information, and iteratively revised scale anchors: 1=poor (largely missing information or unhelpful for viewer use), 2=below average (minimal content with major omissions), 3=moderate (some helpful information but incomplete or unclear for viewers), 4=good (covers main points understandably, though not comprehensive), and 5=excellent (clear, well structured, comprehensive, and highly useful).

Reviewers completed several rounds of independent recoding and consensus review over a few weeks. Disagreements were resolved by consensus using the written anchors and comparison to exemplar videos.

Calibration continued until IRR exceeded 0.90 across all variables and GQS reliability surpassed 0.75. After this, full extraction was conducted, with 1 reviewer coding all 100 videos and the second reviewer independently coding a blinded 10% (10/100) subset. Final IRR for this subset remained high (agreement=97.9%; Scott pi=0.954; Cohen kappa=0.954; Krippendorff alpha=0.956; Multimedia Appendix 2).

Descriptive statistics summarized video characteristics, themes, and creator demographics. Inferential tests included independent samples t tests and one-way ANOVA for comparing continuous variables across groups, Pearson correlations to examine relationships between engagement metrics and quality indicators (ie, accuracy and GQS), and χ² tests for associations between categorical variables. Statistical significance was defined as P<.05. All analyses were conducted in Microsoft Excel and R (R Foundation for Statistical Computing). No videos were excluded due to missing data.

Because engagement metrics (ie, likes and views) were highly skewed, we performed a decile-based stratification. Videos were grouped by engagement deciles, and accuracy patterns were examined across strata to identify nonparametric trends that might have been obscured by extreme outliers.

Analysis of the 100 most-liked TikTok videos generated using the search term “sunscreen” revealed a mean like count of 227,671 (SD 351,902; range 26,700‐2,400,000) and a mean view count of 4,203,243 (SD 5,775,935; range 309,800‐34,300,000). The average number of followers per creator was 1,033,824 (SD 2,977,532; range 272‐18,100,000). Of the 100 content creators, only 18 (18%) identified as health care professionals, including 13 (13%) dermatologists, 2 (2%) medical aestheticians, 1 (1%) pharmacist, 1 (1%) esthetic nurse practitioner, and 1 (1%) cosmetic chemist. Ingredient mentions appeared in 13 (13%) videos, and toxicity concerns in 10 (10%) videos. Positive sunscreen attitudes were expressed in 74 (74%) videos.

Out of the 100 videos, 35 (35%) were classified as accurate, 6 (6%) as inaccurate, 2 (2%) as mixed, and 57 (57%) were opinion-based or not applicable for factual accuracy evaluation. Positive attitudes toward sunscreen were expressed in 74 (74%) videos, with the remainder split between 17 (17%) neutral and 9 (9%) negative sentiment videos.

Sunscreen product recommendations appeared in 85 (85%) videos, and 27 (27%) videos contained explicit promotional content such as affiliate links, brand sponsorships, or paid advertisements (Table 1).

Although inaccurate videos were fewer in number, they received disproportionately high engagement. On average, the 6 (6%) inaccurate videos received 2,141,683 (SD 2,235,616) views and 119,383 (SD 194,427) likes and the 35 (35%) accurate videos received 2,732,594 (SD 3,067,518) views and 135,711 (SD 131,773) likes.

A strong positive correlation was observed between the number of likes and views (r=0.726; df=98; P<.001). In contrast, accuracy was negatively correlated with both likes (r=–0.229; df=98; P=.02) and views (r=–0.242; df=98; P=.02), indicating that less accurate content may attract more user engagement. Refer to Table 2 for all Pearson correlation coefficients among key video characteristics.

To account for the skewed distribution of engagement metrics, a decile-based analysis was conducted. This analysis revealed a strong, statistically significant negative correlation between accuracy and views (r=–0.720; df=8; P=.02) and a moderate, nonsignificant negative correlation between accuracy and likes (r=–0.440; df=8; P=.21). These findings suggest that less accurate videos may be algorithmically favored or more widely shared, resulting in greater visibility despite lower informational quality (Figures 1 and 2).

Content quality was generally low across the top 100 most-liked sunscreen-related TikTok videos. The majority (71/100, 71%) of videos received GQS of 1 (poor quality) or 2 (below average), with only 2 (2%) scoring a 5 (excellent quality). GQS was positively correlated with accuracy (r=0.270; df=98; P=.007), indicating that videos deemed more accurate were also rated as higher quality. However, GQS showed no significant correlation with engagement metrics such as likes (r=0.118; df=98; P=.24) or views (r=0.193; df=98; P=.05), suggesting that content quality did not predict popularity on the platform. These findings reflect a disconnect between perceived usefulness and user engagement.

Promotional content was present in 27 (27%) of the top 100 most-liked sunscreen-related TikTok videos. However, the presence of promotional material was not significantly correlated with either likes (r=–0.161; df=98; P=.11) or views (r=–0.043; df=98; P=.67), suggesting that promotional status alone did not drive engagement. Similarly, promotional content showed only weak, nonsignificant correlations with GQS (r=0.186; df=98; P=.06) and accuracy (r=0.118; df=98; P=.24). These findings indicate that promotional content was common but not necessarily more engaging or accurate than nonpromotional content. Figure 3 visualizes these correlations across 6 content variables.

Our analysis of trending TikTok content on sunscreen revealed an average of 4,203,243 views and 227,671 likes per video, highlighting the platform’s massive reach. Among the 100 most-liked videos, 74 (74%) expressed a positive attitude toward sunscreen, 35 (35%) contained accurate information, and only 6 (6%) contained inaccurate content, per the American Academy of Dermatology guidelines. Notably, 57 (57%) videos were purely opinion-based. Our study is the first to demonstrate that sunscreen content on TikTok is predominantly positive but frequently lacks factual grounding or educational value. These findings expand on prior reports of variable quality in dermatology TikToks by specifically examining sunscreen content, an area of high public interest and misinformation.

Consistent with prior literature, we found that videos deemed accurate were rated significantly higher in quality, as measured by the GQS, aligning with established findings that content accuracy enhances educational utility [3-6]. However, more than half of the videos (57/100, 57%) were categorized as opinion-based and lacked references, reflecting a shift toward anecdotal rather than evidence-based guidance.

Promotional content was present in 27 (27%) videos but was not significantly associated with engagement metrics or content quality. It showed only weak, nonsignificant correlations with GQS (r=0.186; df=98; P=.06), accuracy (r=0.118; df=98; P=.24), likes (r=–0.161; df=98; P=.11), and views (r=–0.043; df=98; P=.67), suggesting that the presence of commercial messaging neither enhanced nor substantially degraded video quality or popularity. Although Ranpariya et al [8] documented the prevalence of promotional content among dermatology influencers, our analysis suggests that such content does not necessarily drive engagement or correlate with accuracy.

Despite comprising only 6 (6%) videos, inaccurate content attracted disproportionately high engagement, receiving a mean of 2,141,683 views and 119,383 likes, which was comparable to that of the 35 (35%) accurate videos (mean views 2,732,594; mean likes 135,711). A strong negative correlation was found between accuracy and views in decile-based analysis (r=–0.72; df=8; P=.02), and a moderate but nonsignificant negative correlation was found between accuracy and likes (r=–0.44; df=8; P=.21). This reinforces prior concerns that social media virality may be driven more by sensationalism or esthetics than by informational accuracy [11-13].

The low overall GQS (mean 1.97) emphasizes the poor educational utility of the majority of sunscreen-related content on TikTok. Despite their reach, most videos failed to meet standards of quality, evidence, or clarity. Content with promotional framing was particularly vulnerable to low-quality scores, even when expressing positive sunscreen messaging. Given that accurate content did not consistently outperform anecdotal or promotional posts in engagement, our findings reinforce the need for dermatologists to adapt communication strategies to the realities of algorithm-driven platforms.

To address these gaps, we recommend that health care providers stay informed about dermatology trends on social media, discuss misinformation and content sources with patients, and engage on platforms such as TikTok to provide evidence-based perspectives.

Encouragingly, 18% (18/100) of the content creators in our dataset were health care providers, and 1 board-certified dermatologist had more than 18 million followers, demonstrating the potential impact medical professionals can have. To further promote quality content, we advocate for standardized disclosure of conflicts of interest, citation of sources, and increased collaboration with platform moderators.

Our team is committed to continuing this work through the SkinMedia series. This initiative aims to systematically track dermatologic misinformation and translate findings into publicly accessible content. This work directly addresses the gap identified in the current literature by moving beyond analysis to the development of interventions that amplify accurate dermatologic messaging. By partnering with dermatologists, educators, and digital platforms, we aim to promote scientifically grounded messaging, particularly for younger audiences. Future work will include the dissemination of open-access educational resources.

This study is limited by its cross-sectional design and the selection of only the 100 most-liked videos, which may not reflect less popular but potentially higher-quality content. The large percentage of promotional content (27/100, 27%) may also have skewed findings. IRR for most variables was high; however, agreement on GQS was lower (κ=0.712), consistent with its known subjectivity. Additionally, most content creators were nonhealth care professionals, which may explain the infrequent use of citations.

This study was one of the first to systematically analyze the most-liked sunscreen videos on TikTok, demonstrating an abundance of opinion-based, low-quality content despite an overall positive stance toward sunscreen use. We urge dermatologists and other health professionals to expand their digital presence to ensure that younger populations have access to engaging, reliable, and evidence-based skin health information. Future efforts should explore collaborative strategies with platforms and influencers to amplify high-quality dermatology content and counteract misinformation.